Your own tuned LLM, running 24/7
A single-tenant managed runtime for local models and agents, operated in Japan. Fixed monthly cost — no per-token bills, no servers to babysit.
Request a pilot slotJoining the waitlist is a plain email — no forms, no account. We only use your address to contact you about pilot onboarding.
The problem
Agents multiply your token bill — every loop they run shows up on the invoice. Self-hosting fixes the bill and hands you a second job: uptime, watchdogs, recovery, model updates. Most teams want the weights and the fixed cost. Not the pager.
What we run for you
We tune an open model to your use case (LoRA), wire it into an agent runtime with monitoring, auto-recovery and human-approval gates, and operate it 24/7 on dedicated single-tenant Apple Silicon in Japan. You get an API endpoint and a simple dashboard — not a server to look after.
What's included
- Use-case LoRA tuning on an open model chosen for your task and its license
- Dedicated single-tenant runtime — your workload never shares tenancy
- Agent runner, monitoring, auto-recovery, human-approval gates for outbound actions
- Model updates and day-to-day operations, done for you
- Your adapters are yours — full export when you leave, our copies deleted
Not included
- SLA — founder-operated, best-effort support during the pilot phase
- SSH or general-purpose compute — this is a managed service, not a server rental
Where your data goes
Standard configuration sends no training data or prompts to third-party AI APIs or cloud GPUs. If a large tuning job needs third-party GPU capacity, we disclose the provider, region, data scope and retention — and proceed only with your written per-job opt-in.
Who's behind this
A solo founder who runs his own company on this exact stack — agents post, monitor and self-heal daily, behind human approval gates. The operations log is published in public.
Pricing
Billed in JPY equivalent. B2B pilot · 3-month initial term · 2 pilot slots. Prices are pilot-phase and will be finalized at general availability.
FAQ
- Which models?
- Open models (Qwen, Llama family and others), chosen for your use case and their license terms.
- Where does my data live?
- On your dedicated single-tenant runtime in Japan. Per-customer isolation for storage, keys and logs; hardware is wiped before any reallocation.
- What happens if I cancel?
- Your adapters and configuration are handed over to you, and our copies are deleted. No lock-in beyond the initial term.
Two pilot slots. We'll reply personally to every request.